Endogenous priors

Estimation ML/Bayesian estimation
1

Hello everyone,

I have a question about the Dynare function computing endogenous priors for likelihood estimation. I understand that it is based on the Christiano-Trabandt-Walentin paper and that it targets variances.

I would be interested in targeting also the first-order autocovariances. I get how to modify the « empirical » part of the code computing data moments. However, I’m not sure how to modify the part computing the theoretical moments. It seems to me that additional inputs to the endogenous_priors function would be needed to do that, but I can’t tell their exact names in the Dynare syntax. Any help from the devs would be welcome.

Also, at the end of the code, there is a commented line corresponding to Del Negro - Schorfheide type VAR priors. It shows the expression of the likelihood, but doesn’t explain how to compute the various elements. I was curious if someone ever wrote a Dynare code to implement this form of endogenous priors.

Many thanks,

A.

2

That seems not straightforward to change at the moment. But you could simply implement this as a system prior as the examples/Gali_2015.mod · master · Dynare / dynare · GitLab example shows.

3

Thanks a lot for your reply, Johannes.

In the end, I think it is easier than I first thought. I passed the Kalman filter transition matrix to the endogenous_prior function, and given Pstar this allowed me to compute the theoretical autocovariances of interest. The code seems to work fine, even tho I would need to cross-check it more.

I will also have a look at your suggestion of using system priors.